iiinova — where your digital self keeps evolving

Bahrad Sokhansanj @bahradx.bsky.social · 07/10/2026

Meta's big selling point was creating individual VMs and model instances for users, but it literally doesn't know how to make a product that isn't anti-social and privacy-violating. It's fascinating.

000

Bahrad Sokhansanj @bahradx.bsky.social · 06/10/2026

We could have come up with what I think is the best possible policy: training for open source models is fair use, and training for for-profit models isn't (and require that private fine-tuners propagate the open source model license). But would rightholders accept that? Based on experience, no.

010

Bahrad Sokhansanj @bahradx.bsky.social · 02/10/2026

Just my personal opinion, but this atrocity exemplifies the reckless culture that has become all too prevalent in tech. Simply no regard for human values. If they stick around, I hope it's only long enough to get sued to oblivion by any victims of frauds perpetrated by this.

000

Bahrad Sokhansanj @bahradx.bsky.social · 01/10/2026

A lot of people still say this is all made up and a scam to hype up OpenAI. There is no doubt that company is spinning like crazy. That's what companies do. But the reality is that inside there are people freaking out. Also, today AG Bonta served a subpoena about recent OpenAi incidents.

DOJ encourages anyone with information regarding this, or any similar cybersecurity incidents or risks, to contact oag.ca.gov/report.

100

Bahrad Sokhansanj @bahradx.bsky.social · 30/09/2026

Once again, time to make it clear that distillation isn't IP theft. To the extent there are security implications or concerns over competition, address those with clear eye to tradeoffs. Companies should protect themselves; the biggest risks here are policy error. See my @lawfaremedia.org article👇

100

Bahrad Sokhansanj @bahradx.bsky.social · 28/09/2026

We can try but will likely fail under the current laws. We should try to pass better ones in the US, Canada, Australia, and everywhere else. I wrote about it here: limitingprinciples.substack.com/p/did-openai...

limitingprinciples.substack.com

When AI Agents Hack, Who Is Liable?

Making the Case that Existing Anti-Hacking Laws Apply Where an AI Agent Selects the Target and Method

140

Bahrad Sokhansanj @bahradx.bsky.social · 26/09/2026

I often see "well the billionaires have decided what AI will be like / shove it down our throats," but people have agency. I also read a lot of "we should throw these guys in jail," but the law doesn't make this easy or perhaps even possible. For that to be a potential consequence, it must change.

010

Bahrad Sokhansanj @bahradx.bsky.social · 25/09/2026

The absence of Trailer Park Boys (or Letterkenny or Kim's Convenience or Orphan Black) from this list is yet another piece of evidence for Canadians that Trump isn't an anomaly, and the relationship with the US has suffered a rupture that goes beyond partisan politics.

020

Bahrad Sokhansanj @bahradx.bsky.social · 24/09/2026

Liability re: open weights is "easier" in some sense than closed source models because the legal relationships are much more similar to other software products and services. That doesn't mean there wouldn't be complications, but it'd more legally familiar than closed/frontier model liability IMHO

120

Bahrad Sokhansanj @bahradx.bsky.social · 23/09/2026

There's a lot of bombastic language here, but the nut of this essay is a call to "take a breath" and pause AI policy. Despite the rhetorical frame, this conclusion in fact favors companies who want to self-regulate without public transparency and accountability.

000

Bahrad Sokhansanj @bahradx.bsky.social · 19/09/2026

For the historical record on the origins of AI safety concerns among US policymakers, I respectfully submit @repbradsherman.bsky.social and his speeches on AI from the mid-2000s. Link to one of them below (couldn't find the 2005 graduation address at UCLA that he referenced in his from this year).👇

100

Bahrad Sokhansanj @bahradx.bsky.social · 18/09/2026

Some real sketchy behavior by Google. Did Irregular withhold the disclosure because Google insisted on it even though at the time there were disclosures of this happening during Irregular tests of Anthropic, OpenAI, and Meta models? Did Google want to look good by comparison?

Irregular notified Google about the hacks at the end of July, Google and Irregular said, in the wake of the discovery that OpenAI's agents had hacked the AI software company Hugging Face. Google didn't disclose the hacks until The Wall Street Journal reached out with inquiries this week.

000

Bahrad Sokhansanj @bahradx.bsky.social · 17/09/2026

I like to make fun of the rationalists and OGers as much as anyone, but this is no longer true even if there ever was a grain of truth in it. A lot of people have found their way to AI safety outside EA (let alone rationalism), and AI safety policy has become professionalized & DC-centric.

1170

Bahrad Sokhansanj @bahradx.bsky.social · 13/09/2026

I'm sure it's fun for OpenAI employees to troll people who support open source AI, but that avoids the relevant question: When will OpenAI tell us why they either concealed the RubyGems attack or failed to investigate it, despite the fact that it should have been fairly easy for them to identify it?

010

Bahrad Sokhansanj @bahradx.bsky.social · 12/09/2026

OpenAI impliedt the Hugging Face breach was the first tand then didn't disclose this for the last two months. Were they lying? Deliberately hiding this? Did they pay Ruby Central to stay quiet? Because it's not like the RubyGems attack wasn't public at the time it happened.

210

Bahrad Sokhansanj @bahradx.bsky.social · 11/09/2026

Nice thread and look forward to the model. It's why I've been highly skeptical of an antitrust exemption for the frontier AI companies to pace/pause, since it's too hard to disentangle safety and commercial objectives.

030

Bahrad Sokhansanj @bahradx.bsky.social · 10/09/2026

It's important to note that despite being outwardly concerned about concentration of power and mass surveillance risks, Dario A. also appeared to endorse military uses of AI (at least as a means of threat) to ensure the superiority of AGI/ASI under "democratic" control in Machines of Loving Grace.

Quote from Machines of Loving Grace by Dario Amodei: My current guess at the best way to do this is via an “entente strategy”26, in which a coalition of democracies seeks to gain a clear advantage (even just a temporary one) on powerful AI by securing its supply chain, scaling quickly, and blocking or delaying adversaries’ access to key resources like chips and semiconductor equipment. This coalition would on one hand use AI to achieve robust military superiority (the stick) while at the same time offering to distribute the benefits of powerful AI (the carrot) to a wider and wider group of countries in exchange for supporting the coalition’s strategy to promote democracy (this would be a bit analogous to “Atoms for Peace”). The coalition would aim to gain the support of more and more of the world, isolating our worst adversaries and eventually putting them in a position where they are better off taking the same bargain as the rest of the world: give up competing with democracies in order to receive all the benefits and not fight a superior foe.

0181

Reposted by Bahrad Sokhansanj

Bahrad Sokhansanj @bahradx.bsky.social · 09/09/2026

If you're interested in reading something that has a more normal take on how policy can respond to the threats of AI, I recommend this (the authors are my colleague but I'd recommend it anyway): radical-optionality.ai

radical-optionality.ai

Radical Optionality — Governing Transformative AI Under Uncertainty

Radical optionality is about preserving democratic governments’ ability to make good decisions about how to govern transformative AI systems as circumstances evolve.

001

Bahrad Sokhansanj @bahradx.bsky.social · 09/09/2026

Yes 100%. We desperately, urgently need a more politically diverse group of people working in AI risk/safety. The kinds of people who are skeptical of this outcome have refused to participate. (This isn't partisan, many of the people who are like this are Democratic-leaning or outright Democrats.)

110

Bahrad Sokhansanj @bahradx.bsky.social · 09/09/2026

This has always been my view. I do think it'd be better to write some precautionary laws and not wait for catastrophes before slowing down and add substantial safety limits on its implementation (i.e. what the sorts of things that let us surviving nuclear power and continue to take a lot of effort).

000

Reposted by Bahrad Sokhansanj

Sung Kim @sungkim.bsky.social · 06/09/2026

When building an agentic AI app, - you do not need memory, - you do not need a router, - you do not need an orchestrator, All you need is a message board!

420719

Bahrad Sokhansanj @bahradx.bsky.social · 03/09/2026

More and more people are saying (that we need to test, and if needed, update the CFAA and state anti-hacking laws to make sure we can hold people accountable for "rogue" AIs). Some initial thoughts I've had in the comments.

100

Bahrad Sokhansanj @bahradx.bsky.social · 03/09/2026

My (admittedly Zitron-esque) take here would be that you wonder whether OpenAI would have declared a previous model release to be "the dawn of the AGI era" if they didn't have that deal with Microsoft that included (now-dropped) contingencies on the achievement of AGI milestones.

1110

Bahrad Sokhansanj @bahradx.bsky.social · 03/09/2026

Right now, efforts to regulate AI are being driven from neoliberal and conservative-libertarian directions. The best proposals rely on private-public partnerships, with an emphasis on for-profit auditors helping regulate for-profit AI companies. It's long past time for progressives to get engage.

000

Bahrad Sokhansanj @bahradx.bsky.social · 03/09/2026

Too few Americans appreciate that while Federalists ended up losing the White House and Congress, they retained control over the Supreme Court and shaped it according to their anti-democratic ideology.

010

Bahrad Sokhansanj @bahradx.bsky.social · 03/09/2026

What happens when AI is causing harm, but there’s no identifiable person to sue or prosecute to stop it? If we don't want governments to limit a free and open AI ecosystem, I think it's time to revisit stronger civil forfeiture and chokepoint authorities. I wrote a rough proposal; link in comments 👇

110

Bahrad Sokhansanj @bahradx.bsky.social · 02/09/2026

I usually like Ben Recht's takes a lot. I'm also highly skeptical of CoT interpretation. But there are 1) factual errors in this thread (Irregular was a separate incident, HF's ability to sue was limited), 2) OpenAI's failure to monitor isn't really a reason to minimize, (1/4)

110

Bahrad Sokhansanj @bahradx.bsky.social · 01/09/2026

People may not know this but Mapquest is a brand that's operated by a company based in Los Angeles (System1).🫶 #LA

010

Bahrad Sokhansanj @bahradx.bsky.social · 30/08/2026

My take on this is that this reflects a legitimate action by a democratically elected US president (within the context of a US system that doesn't care if a treaty refers to a particular name, etc.). We should take what that action implies seriously & if Google resisted that would deny reality.

100

Bahrad Sokhansanj @bahradx.bsky.social · 28/08/2026

Great to see @metr.org post this on Bluesky, and hope that that some of the great people who work there will be more active here and that there will be more AI policy / AI safety engagement on this platform. It's important for engaged progressives to be part of the AI policy discourse.

040

Bahrad Sokhansanj @bahradx.bsky.social · 27/08/2026

Even within the scope of this incident, it appears that other parties were accessed during the "Hugging Face incident."

130

Bahrad Sokhansanj @bahradx.bsky.social · 18/08/2026

I think its important to keep in mind that while their (definitely weird) sense of ethics and internal politics (they have a bunch of people on staff who are into this stuff now) do play a role, the story here is really about social, commercial, and potential legal pressure. 🧵(1/4)

111

Bahrad Sokhansanj @bahradx.bsky.social · 18/08/2026

From personal knowledge there appear to be zero people engaged in AI policy, including staffers, who care about this—certainly to a much lesser degree than hypothetical risks. And yet the impact is real and catastrophic. (And a very good reason to slow data center expansion.)

010

Bahrad Sokhansanj @bahradx.bsky.social · 15/08/2026

Of course such content is an important part of the AI policy discourse, which is why I need to be on the everything app for professional reasons.

020

Bahrad Sokhansanj @bahradx.bsky.social · 11/08/2026

It's well past time to get answers from the big AI companies who are building data centers all over the place as they pursue an unregulated, secretive, and, as has been revealed, increasingly reckless AI race. Congressional hearings on the OpenAI-Hugging Face incident would be a good start.

010

Bahrad Sokhansanj @bahradx.bsky.social · 10/08/2026

I've been thinking, talking, and at times arguing with people about my view that AI companies, like OpenAI in the Hugging Face incident, can be held liable under the CFAA & state anti-hacking laws. It's tough! But I decided to write up the best argument I can for why there could be liability (link👇)

121

Bahrad Sokhansanj @bahradx.bsky.social · 10/08/2026

Everyone mocks Mistral for trailing the race but maybe they're just waiting to sue the leaders.

010

Reposted by Bahrad Sokhansanj

James Fallows @jfallows.bsky.social · 08/08/2026

This by CNN is genuinely reprehensible. Everyone involved should be embarrassed. -The 'new pain' case vs Mamdani is *solely* his religion. -He was in NYC on 9/11, at grade school -Soon after attack GWB himself gave famous speech warning against blaming all Muslims for terrorism -25 yrs later, this.

21169511520

Reposted by Bahrad Sokhansanj

Gautam Hans @gshans.bsky.social · 06/08/2026

I started at Vanderbilt almost a decade ago, to launch their First Amendment clinic. How sad to see how far it has strayed. This policy is a problem six ways from Sunday.

194

Bahrad Sokhansanj @bahradx.bsky.social · 05/08/2026

Questions also around t"Irregular," which OpenAI (per today's announcement) and Anthropic used as a third party evaluator in incidents of models having "inadvertent" internet access. Irregular is a VC-backed for-profit startup (Sequoia is a lead investor and has also invested in the AI companies).

021

Bahrad Sokhansanj @bahradx.bsky.social · 04/08/2026

Gone are the genteel days of quoting from the filing in a PR, I guess.

010

Reposted by Bahrad Sokhansanj

Benjamin Schneider @benschneider.bsky.social · 31/07/2026

Progressive LA Mayoral Candidate Nithya Raman told Vanity Fair that her main difference with DSA is on housing. www.vanityfair.com/story/nithya...

2777397

Bahrad Sokhansanj @bahradx.bsky.social · 31/07/2026

The context though that they basically took out all of the internal use stuff from their SB53 / EU AI act frameworks. All this stuff is in the RSP, for which there is no legal enforcement. So even if they address this in their policy it doesn't mean anything because it isn't a legal obligation.

011

Reposted by Bahrad Sokhansanj

mr. TIM @timkellogg.me · 30/07/2026

hear me out: slow down development of the highest performing frontier models, and instead just drive costs down

7441

Bahrad Sokhansanj @bahradx.bsky.social · 30/07/2026

The OpenAI/HF incident urgently shows the potential risks of both training and deploying long-horizo... Read more: longer.blue/posts/1D7ae-hbJO

000

Reposted by Bahrad Sokhansanj

Gautam Hans @gshans.bsky.social · 29/07/2026

Teaching PR radicalized me into an abolish the bar person. This latest fiasco proves that states can’t even be trusted to administer a pointless exam. Cancel it & redirect the funds towards ethics compliance. Hard to pull off given entrenched interests & the cartel nature of the bar. But we can try!

15610

Bahrad Sokhansanj @bahradx.bsky.social · 25/07/2026

This (and the thread as a whole) is an important insight about Anthropic's thinking and about people misinterpreting sincerity for commercial ambition. Although when you're thinking about regulation, motives are usually immaterial; it's the conduct that counts.

020